Published on September 13, 2026, the brief describes a Transformer with a recurrent decoder that carries state between tokens during the prompt and generation.
Published on September 13, 2026, the brief presents the Recurrent Looped Transformer (RLT). Its causal encoder creates a global KV memory; the recurrent decoder also uses a KV cache of recent tokens and the decoder's final state from the previous token. This state persists through the prompt and during generation.
According to the brief, this approach extends a Transformer's temporal information path without increasing the decoder blocks per token. To check the details, consult the original publication or project material and compare its account of the architecture and reported results with this summary; the supplied text specifies no metrics or evaluations.